Every company is sitting on a goldmine. Google knows this and mines it. Microsoft knows this and mines it. Why not mine it yourself?
Customers mention things casually all the time:
“Geez, I’m slammed this week, I have to get around to a dozen production sites!”
“Juggling timezones is hard, we’ve got suppliers and customers in [list of cities around the world].”
“That day’s kinda busy, my kid’s graduating in the evening so we’re setting up a little surprise party.”
That is your company’s email – millions of messages back and forth with customers, full of useful detail that probably never got entered into your CRM by the sales team. Assuming your CRM even has places for it.
Unless you are one of those superhuman people with an eidetic memory, it is impossible to keep all of that at your fingertips across thousands of contacts.
Why not just buy a tool for it
These are sensitive details, and you do not want a situation like Samsung, where sensitive information leaked out through AI tools – or ChatGPT exposing conversations to search engines.
There are plenty of tools for extracting this data already. They also:
- literally require access to all of your historical information just to do the work
- have every reason in the world to monetise that information themselves
- have an unknown security profile – look at the exceptions in any SOC 2 report these days
- carry a large price tag
Even with your own private AI enclave you would still need to clean up the data, decide what to extract and how, work out where to store all those small details, and then work out how to use them once you have them.
Use AI to solve the problems AI created
Use a model to expand your list of things to look for. Then have it write scripts that:
- extract only the customer emails, skipping vendor newsletters and the rest
- anonymise them – strip older quoted sections and signatures, and replace names, numbers and other identifying details with placeholders
- hand those up, with an appropriate prompt, to a large hosted model
- receive the results, de-anonymise the useful details, and store them locally
The key point is that the sensitive data never leaves in a form anyone can use. The heavy lifting happens on anonymised text.
It sounds like a lot of work. With AI assistance, internal IT staff could do it in less than a day – about the time it would take to set up a third-party service – and for pennies on the dollar, with far greater control over the process, the data and the budget.
A note on which model writes the scripts
The good thing about having AI build tools like this is that you do not have to deal with the nitty-gritty. Your problems are not unique; they are solved problems many times over and the code already exists.
Are there better and worse choices for this? Yes. Some services will happily suggest application libraries that either do not exist, or have been created by bad actors who noticed those names were being hallucinated. Others are better about it, though none is perfect.
Out of the hundreds of thousands of lines of code and the dozens of libraries pulled in across various projects, I have yet to find one that was not the bog-standard, mainstream, widely accepted, best-in-class option. The same goes for database engines, storage structures and file formats – solved problems, with no need to spend developer hours reinventing the wheel.
Now flip it around and write the emails
Everything above applies just as well to new mail as to the historical archive – watch for incoming customer email, anonymise, extract, de-anonymise, store. That is simply keeping up with what arrives.
But now that the extraction has happened, why not draft a reply? The next scripted step:
- retrieves the sender’s profile of known data points
- anonymises the sensitive parts – the child’s name becomes a placeholder
- sends the anonymised profile and the anonymised email, with a prompt for generating a reply, to the external service
- de-anonymises the response and saves it as a draft
It will not be a perfect, complete reply. But a personalised response that only needs a few words changed – in a tone that follows the pattern of previous conversations – saves a minute or more on every email without looking like a template. Or like AI.
Do that arithmetic across the hundreds of emails a busy salesperson sends in a day and it is a great deal of time for a minimal investment and a tiny fraction of the recurring cost.
Itwerx Corp is a service-disabled veteran-owned small business providing IT services across Seattle, Bellevue, Everett and Snohomish County. This is the kind of thing our business process analysis work deals with – talk to us about yours.
